A $400 Million Deal Reshapes the AI Infrastructure Map

A New Era in AI Financing

In a striking move that reflects the rapid shifts in the AI infrastructure market, General Compute — a startup specializing in cloud inference services — has announced a $400 million loan from technology investment firm Upper90. What makes this deal exceptional is not just its size, but its nature: it is the first of its kind to use specialized inference chips as loan collateral. These chips are purpose-built to run pre-trained models with high efficiency and low cost, as an alternative to the expensive graphics processing units (GPUs) typically used during the training phase.

Why Inference, Specifically?

Growing concerns in tech circles about the costs of AI tools and tokens have pushed investors toward infrastructure capable of running open-source models at lower costs compared to large language models from leading labs. This is where inference chips come in, enabling access to AI capabilities without the heavy spending required for traditional infrastructure.

The company was founded by CEO Finn Bukovszky and CTO Jason Goodison, and raised $15 million in a seed round last May to build a specialized inference cloud powered by chips from Intel-backed SambaNova. Companies of this type are known as "Neoclouds" — providers that focus exclusively on AI workloads, in contrast to the general-purpose infrastructure offered by traditional giants like AWS and Azure.

The Advantages of SambaNova SN50 Chips

The SN50 chips possess technical characteristics that make them a serious competitor in the inference market:

• Superior power efficiency compared to traditional GPUs.
• No requirement for costly liquid cooling systems, accelerating deployment.
• Flexible distribution across diverse data centers.
• Inference speeds up to 16 times faster than GPU-based clouds, according to the company.

From GPUs to Inference Chips: Upper90's Journey

This is not the first time Upper90 has ventured into financing specialized chips. Co-founder and CEO Billy Libby — a former quantitative trader at Goldman Sachs — recounts how his firm was among the first to finance GPU purchases in 2021 for Crusoe, a data center company focused on energy efficiency. It was a historic move that traditional banks were unwilling to make at the time, due to concerns about chip depreciation over time.

What began as a high-risk venture has since become an established business model, as CoreWeave transformed chip-backed loans into a cornerstone of its successful IPO. Libby reflects on this shift: "When we financed Nvidia chips as the first group to do so, the market was inefficient — we were getting real compensation for the risk." Today, with GPUs increasingly saturated in the market, the firm is turning its focus to the next wave: inference chips.

A Growing Openness to Nvidia Alternatives

This deal comes amid a broader trend of rising interest in technical alternatives outside the Nvidia ecosystem. Companies like Groq and Cerebras are attracting growing attention from acquirers and financial markets, while TensorWave is betting on an AMD partnership to deliver alternative infrastructure. Platforms providing access to open-source models — such as OpenRouter and Fireworks — have also closed funding rounds at high valuations, and new models like Kimi K3 have demonstrated the ability to compete with the latest releases from Anthropic and OpenAI in coding benchmarks.

Bukovszky sums up the big bet behind this deal: "There are chips that are beginning to scale and deliver excellent total cost of ownership or speeds that surpass Nvidia, but buyers are still few. This deal isn't just a startup that got funding to buy some compute — it's the first signal that capital is organizing and that Nvidia's monopoly hold is beginning to crack."

Conclusion: Inference Is the Next Battleground

In a world where everyone is racing to build smarter, more powerful models, it appears that smart capital is placing its real bet on the other side of the equation: running those models efficiently and at a reasonable cost. The General Compute and Upper90 deal is more than just an investment story — it is an indicator of shifting market priorities, where not everyone needs a supercomputer, but everyone needs inference and AI.

✦ بقلم فريق دروب أيديا